By Ben van der Schaaf, Partner, and Ben Enejo, Partner, Healthcare & Life Sciences Practice, Arthur D. Little
Twitter: @adlittle
Much has been written about the application of Artificial Intelligence and Machine Learning (AI/ML) in clinical development, and at first glance it is an obvious area of focus. Successful use of AI/ML capability requires vast amounts of data, and there have been more than 400,000 clinical studies registered globally since the beginning of the century. This number is growing at a good pace, generating a very considerable data set across many therapeutic areas and patient cohorts. In practice it has not been that straightforward. The data is of course widely dispersed, of mixed and often questionable quality and organizations’ capability to use this data effectively has been limited, to put it mildly. That doesn’t make it any less relevant, and the drive toward bringing AI/ML capability into the clinical trial space is picking up speed. It is becoming increasing clear that remaining competitive in the space without it will be a futile exercise.
Why is AI relevant for clinical trials?
The number of potential use cases is significant. AI can be used to enhance and accelerate enrollment, the number one challenge in clinical trial operations. This is done by improving the selection of countries and sites with the highest probability of patient enrollment success, by being able to pre-identify patients using EHRs and other data sources and by generally using broader data sources than just the company’s own clinical trial history.
While enrollment is the number one challenge, it is far from the only one, and effective use of AI/ML can help predict quality issues (another major pain point), support monitoring and query resolution, enhance scenario development and protocol design analysis, for example.
Leveraging real world data (RWD) is another high potential avenue to go down, and all of these have in common that they can shorten the time to market (or to trial termination) which is worth a lot to patients, the medical community and drug developers.
The obvious question is then why pharmaceutical companies and CROs are not all over this? In fact, we would say they are, but a number of factors get in the way:
- Clinical operations professionals are first and foremost focused on their ongoing portfolios. They are generally reluctant to experiment and adopt with new ways of working that may delay the next milestone
- Data is literally all over the place. The larger trial sponsors and CROs have very large volumes of data, but it is in many different places, formats, and systems – even within the same company
- There are not enough data scientists and general AI/ML capability is lacking
- The challenge it not contained within an organization – it involves the wider ecosystem
- It will require significant resources for an extended period. This is not a quick fix
All these barriers can be addressed. It will require action at company level – and within the clinical operations ecosystem. At the same time, addressing this is not optional for a company that wants to remain competitive.
When not if!
The journey must begin with recognizing the value of and need for addressing this, and appreciating the complexity of the challenge. To start with the value: companies spend billions every year on clinical trials, and the patent end date of a drug is fixed. Being able to accelerate trial outcomes and increase the likelihood of a positive outcome will drive drug development costs down and generate better outcomes for patients, faster. Regarding the complexity, this is not an easy challenge to address, but it also does not need to be fixed all at once. There does need to be some focus and a larger plan though. Right now, many organizations have initiatives ongoing to introduce AI/ML capability into their clinical trials, but they are fragmented and not connected to systemic efforts that are sponsored at the most senior levels. In most cases these efforts do not go beyond the company boundary to involve the many other players in the ecosystem (CROs, sites, and many other services providers). This is inefficient and will result in marginal improvement at best.
Conclusion
It is easy to see why clinical operations is late to the AI/ML party. There is always a trial that needs to be enrolled and completed and we can’t take any risks with the timeline, so not in my trial please. At a company level, it may not be the sexiest area to start playing with AI/ML either. Right now, drug discovery/research and commercial are the two areas within pharma where most efforts are focused. On top of that, the data and systems infrastructure combined with the lack of internal capability in most companies makes most executives think twice before taking on the challenge. However, it needs to get done. For many it will be a multiyear effort, and it may at times be painful, but there is no doubt that not doing it will be more painful, if not fatal, in the long run. Develop a roadmap, get organized, and get started. It will be worth it.